Research On Network Intrusion Detection System Based On Deep Belief DBN-Granular Computing

With the popularization and development of the Internet and artificial intelligence,the Internet became the medium of the Internet of all things,and it provide a strong impetus for the development of the world.For the Internet,people share the unlimited convenience and experience,but they also face threats and losses due to malicious network intrusion attacks.Improving the effectiveness of existing network intrusion detection systems,mining unknown network intrusion capacity,and improving network information security technology are urgent problems to be solved.Among them,the research on algorithm efficiency and detection accuracy is the focus of breakthrough in this field.In this thesis,the popular deep learning and granular computing theory is applied to the network intrusion detection system.By improving some defects of the traditional algorithm,the effect of the existing network intrusion detection system and the ability of mining unknown network intrusion behavior are improved.In the construction of intelligent intrusion detection system architecture,two processing modules of data feature abstraction and trained and intelligent network agent behavior judgment engine are studied and designed.Aiming at the data dimension reduction,DBN based on BP error feedback was used to transform high-dimension feature vector into low-dimension vector while preserving the features of data sets.In order to improve the intrusion detection efficiency of DBN network and find the characteristics of small-scale unknown abnormal network,combined granular computing theory with clustering algorithm and proposed DBN-GRC model,which can quickly and accurately discover the unknown abnormal traffic through efficient clustering algorithm.In this thesis,the improved CIC-IDS-2017 data set is used as the sample test data set.Through the experimental analysis,the DBN-GRC model is obviously better than the general DBN model.In the detection of small-scale intrusion with known features,it can improve the detection efficiency by 2%-3%,and increase the ability to detect unknown abnormal traffic(the detection rate of unknown anomaly in sample data set is about 0.2%),which achieves relatively good anomaly detection effect and effectively fills the detection function blank of DBN model.